The tarball contains one .txt file per object. The line-based text file can be parsed as follows:
line 1: object id line 2: object name line 3: object description line 4: estimated purity (fraction of flickr query results that actually contain the object) line 5: image id, image width, image height, list of bounding boxes (x1:y1:x2:y2; x1:y1:x2:y2), image url
As an example, the pineapple.txt file looks like this:
32 pineapple whole pineapple fruit (with skin on) 0.037641 3271801, 500, 500, 70:50:252:406, http://farm5.staticflickr.com/4136/4810676140_85fd7e784e.jpg 3223927, 677, 453, 452:1:225:330, http://farm8.staticflickr.com/7146/6786344459_decbeafa89_o.jpg 3274868, 680, 451, 17:0:612:414, http://farm5.staticflickr.com/4070/4705967090_ddc2eb8f06_o.jpg 3223983, 480, 640, 103:20:209:504, http://farm8.staticflickr.com/7148/6784649289_a84343d4c6_o.jpg 3241033, 465, 660, 6:4:200:225;370:241:92:253, http://farm7.staticflickr.com/6051/5913796662_1fbe18e530_o.jpg ...